Swahili words from a school in Mombasa, each placed on a map by its vector so related words sit together. A word the model never saw, "wanafunzi" (students), arrives: a word-level model drops it at <unk>, while FastText builds a vector from its character pieces and lands it beside "mwanafunzi". Then use one-letter pieces, or train GloVe on raw counts and watch "na" pull everything in. This lesson needs JavaScript.